The goal of the Generic Code Optimization (GCO) project is to develop a framework for empirical performance optimization. Many tuning systems achieve good performance in part because the algorithms to be optimized are known ahead of time, so problem-specific techniques can be applied. In our research, we would like to address this limitation by applying techniques similar to those used in problem-specific tuning systems to the optimization of arbitrary code. Since the algorithm to be optimized is not known in advance, it will require compiler technology to analyze the source code and generate the candidate implementations. The code generator/compiler would then be combined with infrastructure for performing the search, generating timing drivers, Makefiles, etc. The combination of all these components into a coherent automatic tuning system would be the ultimate objective of the GCO project.